the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 05 Aug 2026
Magnitude and controls of snow sublimation at a high-elevation Swiss Alpine site
Abstract. Surface snow sublimation remains poorly quantified in the European Alps, with previous modeling studies estimating a broad range of winter surface sublimation losses (10 to 180 mm w.e.). This creates large uncertainties in the partitioning of snow ablation between sublimation and snowmelt, with significant implications for downstream water availability, as surface mass losses through sublimation directly reduce spring meltwater availability. Field observations used to quantify surface snow sublimation throughout a winter season remain scarce due to the logistical challenges posed by high-elevation Alpine environments. To address this, we estimated surface snow sublimation using the eddy-covariance method over one full winter season (November 2024 to June 2025) at the Weissfluhjoch research site (2536 m a.s.l) in the Eastern Swiss Alps by measuring water vapor fluxes with an integrated open-path gas analyzer and sonic anemometer (IRGASON). Partial correlation analyses and explainable machine learning (XGBoost model with Shapley Additive Explanations) were used to quantify the relative importance of different meteorological variables on modeled sublimation. Partial correlation analyses and explainable machine learning (XGBoost model with Shapley Additive Explanations) were used to quantify the relative importance of different meteorological variables on modeled sublimation. Over the 2024–25 winter season, cumulative net surface sublimation was 24.7 ± 18.3 mm, equivalent to 5.7 ± 4.2 % of maximum snow water equivalent, with an average daily sublimation rate of 0.15 mm d−1. Nearly half of cumulative surface sublimation occurred after peak snow height was reached, between March 31, 2025 and June 1, 2025. Vapor pressure gradient and wind speed emerged as the dominant controls on surface sublimation across both statistical and machine learning approaches, while net shortwave radiation became increasingly important following peak snow height. In contrast, air temperature and snow surface temperature showed little independent predictive power once covariance with other variables was accounted for. Our findings highlight the importance of continuous eddy-covariance measurements for quantifying snow sublimation and provide one of the first winter-season estimates of surface snow sublimation in the European Alps.
- Preprint
(9844 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-4268', Anonymous Referee #1, 15 Sep 2026
-
AC1: 'Reply on RC1', Harsh Beria, 02 Oct 2026
Referee #1
Isabella Anglin, on behalf of all authors
I have reviewed the paper “Magnitude and controls of snow sublimation at a high-elevation Swiss Alpine site” by Anglin et al. for publication in The Cryosphere. The study presents results from a field-based experiment at an alpine study site in the European Alps where eddy covariance instrumentation was combined with meteorological and snow monitoring data to quantify surface snow sublimation fluxes over one winter season. Furthermore, the dominant controls on surface sublimation fluxes were assessed using partial correlation and machine learning approaches. Results highlight vapor pressure deficit and wind speed as the dominate drivers of sublimation losses and an average snow season sublimation rate of 0.15 mm day-1 with greater fluxes occurring following peak snow height. I believe this paper addresses an important topic and clearly demonstrates a contribution to the field that is relevant to the readership of The Cryosphere. The paper is clearly organized and well written overall. My comments on the paper are outlined below:
We thank referee #1 for a positive evaluation of our study, we have replied to your comments below.
General comments:
1) One limitation of the current study is that only one year of data are available for analysis. So, the natural question is: what would we expect the year-to-year variability of sublimation fluxes to be with respect to the one year of data presented? The authors do a nice job of highlighting results from previous studies in the discussion relevant to interannual variability of sublimation. Furthermore, they put the 2024-25 winter in context from a snow accumulation and snow cover duration perspective. However, given the history of met data collected at this site, did you consider looking at some of the historical data of the identified drivers of sublimation (e.g., VPD, wind speed, radiation) along with snow conditions to provide a historical context? Can the controls of sublimation be used to provide more information about how sublimation fluxes may vary? At a minimum, I think more discussion around this topic and how future research can benefit from quantifying the controls of sublimation would be helpful for the paper (i.e. expand on why it is helpful to identify the controlling variables of sublimation).
Answer: Indeed, a limitation of this study is that we only have one year of data available. We will therefore put the meteorological conditions observed during the 2024-25 winter in context to the long-term climatology of the site and discuss potential implications on surface snow sublimation.
2) Recent papers including this one have highlighted well that sublimation fluxes become larger and more important as the snow season progresses with the largest sublimation fluxes occurring presumably during the snowmelt period. It would be helpful to provide snow pit and/or snow scale SWE on your depth graphs to provide more context on when the snowmelt season began for this study. The presence of liquid water in the snowpack and particularly near the surface may be an important condition promoting sublimation. Looking at Figure 2 and Figure A1 it appears that there was still a substantial snowpack on 1 June and the analysis removed most of the active snowmelt period following 1 June. While the authors present clear logic for this period of analysis, it would be helpful to comment on how much more sublimation may have occurred during those last few weeks of snowmelt when substantial sublimation rates may have been occurring. Also, further to this point, Table 2 does not highlight a substantial difference in the controls of sublimation (VPD, wind) between the before peak HS and after peak HS period which I would think is another important point of discussion in the paper.
Answer:
- We chose a rather strict definition for the winter period to ensure that transpiration and soil evaporation from exposed ground could be excluded. However, it would indeed be interesting to also quantify the sublimation rates in the last weeks of snowmelt. We will report the latent heat flux during this period in the revised manuscript. However, we will not be able to quantify how much of these latent heat fluxes come from transpiration / soil evaporation / sublimation.
- We will discuss periods with similar VPD and wind speeds before and after peak HS more extensively in the revised manuscript.
3) Please comment on how you confirmed that only surface sublimation fluxes were being measured and quantified in this study and that measured fluxes do not include blowing snow sublimation from saltation and/or near surface suspension of blowing snow? Although the authors mention the lower wind speeds at the study site compared to other studies, wind speeds do appear to approach 10 m s-1 (Figure 2). Blowing snow can often obscure the sensor path of IRGASON eddy covariance sensors so it would also be helpful to provide comments on how blowing snow may have contributed to the need to gap filling.
Answer: Given the sheltered location of the study site and the typically low average wind speeds, we do not expect blowing / drifting snow sublimation to be a dominant process. A study by Groot Zwaaftink et al. (2013) simulating drifting snow sublimation at the nearby Wannengrat in the Swiss Alps (approximately 2 km from Weissfluhjoch research site) estimated that seasonal drifting snow sublimation accounts for only 0.1% of winter precipitation. We therefore expect this process to have minor effect on total sublimation. We will acknowledge this in the revised manuscript. In addition, we do not have any sensors to measure blowing snow sublimation at the Weissfluhjoch research site.
Specific comments:
Line 90: change “fall” to “falls”
Answer: Thank you for catching this. We will adjust it.
Lines 90 – 91: Can you provide more detail or perhaps a wind rose diagram highlighting how common winds from the northwest versus the south are at the site? Which condition is more dominant?
Answer: We will provide more context and add a wind rose diagram to highlight the common wind direction in the appendix.
Lines 97 – 99, Table 1: Is the 10 min and 2 min measurement frequency for wind speed, radiation, etc. based on an average of a more frequent scan interval by the datalogger (e.g., 15 seconds), or do the meteorological sensors truly only take one measurement every 10 minutes?
Answer: These are instantaneous values. We will mention this in the revised manuscript.
Lines 111 – 113: Data gaps – can you comment more on the frequency and length of data gaps that were required to be filled in this study and if there were periods of missing data where simple linear interpolation was not appropriate and rather implementing and diurnal pattern was needed. Were there any thresholds used for the length of missing periods and what techniques were used?
Answer: We will provide details about the frequency and length of gaps for each variable in the appendix of the revised manuscript.
Lines 142 – 143: How were the 10 minutes met data converted to the 15 minute EC flux interval. Also, later used in the partial correlation and XGBoost analysis?
Answer: Linear interpolation was used, given the meteorological measurements are instantaneous. We will mention this in the revised manuscript.
Equation 4: Does this logic make sense to include in the paper since the snow surface temperature was capped at 0 deg C for the data presented (lines 118 – 120)?
Answer: At values of 0 deg C, the specific latent heat is different compared to below 0 deg C (see equation 4), hence it is included. We will revise Eq. 4 accordingly.
Lines 157 – 158: Please provide more detail on the random forest gap filling approach used in this study. Which variables were used as predictors? Also, either here or in the results section, it is important to note how much gap filling was required over the season and during what time periods. Are there certain meteorological conditions that require gap filling? What about conditions during blowing snow? Although the authors briefly mention gap filling in the discussion, more detail is needed in the methods and results.
Answer: While we referenced the DIIVE toolbox used for gap-filling in the current manuscript, we will provide more details about the gap filling procedures in the revised manuscript.
Lines 212 – 216: Did you consider using vapor pressure of the atmosphere only in your analysis as a separate predictor? Similarly, did you consider using temperature gradient between the atmosphere and snow surface?
Answer: We indeed considered these variables separately as reported in Figure A2, we will improve the description in the revised manuscript.
Lines 244 – 249: If you prefer not to present the SSG SWE data, can you plot the SWE observations from snowpits along with your snow depth plots? It is important to try to better highlight potential differences between peak snow height and peak SWE, and more specifically when the snowpack was melting. Also, was liquid water content collected in the snowpits? If so, you could also consider also showing those datasets to further strengthen understanding of when the snowpack was melting and how those periods related to sublimation fluxes.
Answer: We will include the manual SWE measurements and lysimeter measurements in Figure 2 for more context and discuss the potential uncertainties from the definition of peak snow depth vs peak snow water equivalent in the revised manuscript. Unfortunately, we do not have liquid water content measurements during the study period.
Lines 269 – 271: Consider if this is the best way to phrase this sentence about the residual term. Although you don’t directly measure internal energy changes, in theory you measure to processes driving these internal energy changes during the snow accumulation period when melt/freeze is not occurring.
Answer: We will rephrase this sentence.
Lines 312 – 315: Please include more detail on the gap filled periods.
Anwer: We will include information on gap-filling periods in the methods.
Figure 5: Consider if there is another way to label cumulative sublimation in the legend and caption to avoid confusion that the plot is presenting only gap filled values.
Answer: We will adjust the caption to improve clarity.
497 – 503: Surprising to see the Stössel et al., 2010 citation being introduced this late in the discussion considering it was conducted at the same study site. Can results from this previous study be more directly compared to the results presented in this work?
Answer: The study by Stössel et al. (2010) primarily focused on surface hoar formation but also measured mass gain and loss during two periods each for one week. Thus, the temporal scale is different compared to our study, we will incorporate the reference earlier in our discussion of the revised manuscript.
506 – 510: Can you comment on if the unmeasured internal energy changes (i.e. residual term) presented in this study is a realistic magnitude based on melt/freeze cycles and internal snow temperature changes?
Answer: We will compare to literature that directly measures the internal energy change (Helgason & Pomeroy, 2012; Lackner et al., 2022) and comment on this in the revised manuscript.
References:
Groot Zwaaftink, C. D., Mott, R., and Lehning, M.: Seasonal simulation of drifting snow sublimation in Alpine terrain, Water Resour. Res., 49, 1581–1590, https://doi.org/10.1002/wrcr.20137, 2013.
Helgason, W. and Pomeroy, J.: Problems Closing the Energy Balance over a Homogeneous Snow Cover during Midwinter, Journal of Hydrometeorology, 13, 557–572, https://doi.org/10.1175/JHM-D-11-0135.1, 2012.
Lackner, G., Domine, F., Nadeau, D. F., Parent, A. C., Anctil, F., Lafaysse, M., and Dumont, M.: On the energy budget of a low-Arctic snowpack, Cryosphere, 16, 127–142, https://doi.org/10.5194/tc-16-127-2022, 2022.
Stössel, F., Guala, M., Fierz, C., Manes, C., and Lehning, M.: Micrometeorological and morphological observations of surface hoar dynamics on a mountain snow cover, Water Resources Research, 46, https://doi.org/10.1029/2009WR008198, 2010.
Citation: https://doi.org/10.5194/egusphere-2026-4268-AC1 -
AC2: 'Reply on RC1', Harsh Beria, 02 Oct 2026
Publisher’s note: this comment is a copy of AC1 and its content was therefore removed on 5 October 2026.
Citation: https://doi.org/10.5194/egusphere-2026-4268-AC2
-
AC1: 'Reply on RC1', Harsh Beria, 02 Oct 2026
-
RC2: 'Comment on egusphere-2026-4268', Anonymous Referee #2, 18 Sep 2026
This manuscript presents really interesting results from a seasonal study of sublimation measured through eddy-covariance. Using higher level statistics and ML methods, the study provides a novel view into the drivers of sublimation. They show that the vapour pressure gradient (VPG), wind speed, and net short wave radiation are the strongest drivers of sublimation. Additionally, they show interesting behaviour using SHAP analysis that demonstrated a plateau in the control by VPG, and also that SWR becomes more important in the later season. This manuscript is of high interest to the community!
Major Comments:
- I am concerned that choosing the two periods based on the peak snow depth is not a robust enough characterization of the snowpack as it is dependent, at a seasonal level, on individual weather patterns and therefore will exhibit a lot of inter-annual variability. If it wasn’t for one weather event, the peak snow depth may have easily been in late January in 2025. For example, while looking at Fig. 2, I see a plausible three-phase structure: development, maintenance, and collapse of the snowpack, which aligns with three phases in the net radiation, as we start seeing a linear increase in positive net radiation starting February 2025. The third phase, when the net radiation has increasing variability, starts in April 2025. Additionally, as mentioned in the discussion, sublimation is not dependent on snow depth, meaning the peak snow depth should not be used in characterizing sublimation. Can you check previous year snow depth and radiation data or the literature for a more robust characterisation of the snowpack?
- The main machinery of this work is statistics, yet I would like to see a bit more on the physics side. Monin-Obukhov Similarity Theory (MOST) is mentioned at a couple relevant points, how much of the statistics and ML methods tells us anything beyond the MOST bulk relations (VPG and wind speed fitting directly into MOST)? In other words, to what extend is the statistics rediscovering MOST? What level of variance cannot be explained by MOST?
- Late in the season the snow surface temperature stays near 0 deg and therefore stops varying, so it no longer reflects changes in the energy input. In this period, is SWR genuinely driving sublimation, or is it the only varying predictor of energy availability? This could be tested by seeing if SWR remains a significant predictor if the cases where the surface is at 0 deg are excluded.
Minor Revisions:
Line 30/75 – Would measuring sublimation with an IRAGSON also include blowing snow sublimation? Please make clear that for the text sublimation means surface sublimation after the introduction. Additionally, when considering statistically the role of wind in surface sublimation, there might be a contribution from blowing snow.
Line 130 – Why did you only choose 10:00-17:00 for the MRDs? Stable conditions at night might require a smaller averaging window than 15min (1min and 5min are common in the literature). This choice may result in the removal of too much, potentially good, data for stable conditions.
Fig 3 – It is difficult to really see trends due to the bar plot, black and blue are hard to distinguish. Can you try curves instead of bars to see if it improves readability (or some other solution)?
Fig 4(a,b) – The radiative energy flux does not match the sign convention, changing the outward fluxes to -values or everything into absolute values will result in the current plot.
Fig 4(d) – Why is there so much asymmetry in the diurnal cycle of sensible heat flux?
Citation: https://doi.org/10.5194/egusphere-2026-4268-RC2 -
AC3: 'Reply on RC2', Harsh Beria, 02 Oct 2026
Referee #2
Isabella Anglin, on behalf of all authors
This manuscript presents really interesting results from a seasonal study of sublimation measured through eddy-covariance. Using higher level statistics and ML methods, the study provides a novel view into the drivers of sublimation. They show that the vapour pressure gradient (VPG), wind speed, and net short wave radiation are the strongest drivers of sublimation. Additionally, they show interesting behaviour using SHAP analysis that demonstrated a plateau in the control by VPG, and also that SWR becomes more important in the later season. This manuscript is of high interest to the community!
We thank referee #2 for a positive evaluation of our study. We have replied to the individual comments below.
Major Comments:
- I am concerned that choosing the two periods based on the peak snow depth is not a robust enough characterization of the snowpack as it is dependent, at a seasonal level, on individual weather patterns and therefore will exhibit a lot of inter-annual variability. If it wasn’t for one weather event, the peak snow depth may have easily been in late January in 2025. For example, while looking at Fig. 2, I see a plausible three-phase structure: development, maintenance, and collapse of the snowpack, which aligns with three phases in the net radiation, as we start seeing a linear increase in positive net radiation starting February 2025. The third phase, when the net radiation has increasing variability, starts in April 2025. Additionally, as mentioned in the discussion, sublimation is not dependent on snow depth, meaning the peak snow depth should not be used in characterizing sublimation. Can you check previous year snow depth and radiation data or the literature for a more robust characterisation of the snowpack?
Answer: The manual SWE measurements are conducted only once every 2 weeks, therefore at a much lower temporal resolution than other variables, which limits a more robust characterization of different snow periods based on SWE. We will, however, put the meteorological conditions observed during 2024-25 winter in context to the long-term climatology at the site, and evaluate additional data and literature.
- The main machinery of this work is statistics, yet I would like to see a bit more on the physics side. Monin-Obukhov Similarity Theory (MOST) is mentioned at a couple relevant points, how much of the statistics and ML methods tells us anything beyond the MOST bulk relations (VPG and wind speed fitting directly into MOST)? In other words, to what extend is the statistics rediscovering MOST? What level of variance cannot be explained by MOST?
Answer: We will further discuss the implications of our findings in relation to MOST theory in more detail in the discussion.
- Late in the season the snow surface temperature stays near 0 deg and therefore stops varying, so it no longer reflects changes in the energy input. In this period, is SWR genuinely driving sublimation, or is it the only varying predictor of energy availability? This could be tested by seeing if SWR remains a significant predictor if the cases where the surface is at 0 deg are excluded.
Answer: Thank you for this important comment. In the revised manuscript, we will evaluate if SWR remains significant when the snow surface temperature at 0 deg C is excluded.
Minor Revisions:
Line 30/75 – Would measuring sublimation with an IRAGSON also include blowing snow sublimation? Please make clear that for the text sublimation means surface sublimation after the introduction. Additionally, when considering statistically the role of wind in surface sublimation, there might be a contribution from blowing snow.
Answer: Given the sheltered location of the site and the generally low average wind speeds, we do not expect blowing / drifting snow sublimation to be a dominant process measured by the IRGASON. A study by Groot Zwaaftink et al. (2013) simulating drifting snow sublimation at the nearby Wannengrat in the Swiss Alps (approximately 2 km from the Weissfluhjoch research site) estimated that seasonal drifting snow sublimation accounts for only 0.1% of winter precipitation. We therefore expect the effect of this process on sublimation totals to be small. We will acknowledge this in the revised manuscript. In addition, we do not have any sensors to measure blowing snow sublimation at the Weissfluhjoch research site.
We will also improve the definition of the term sublimation as used in our manuscript directly after the introduction.
Line 130 – Why did you only choose 10:00-17:00 for the MRDs? Stable conditions at night might require a smaller averaging window than 15min (1min and 5min are common in the literature). This choice may result in the removal of too much, potentially good, data for stable conditions.
Answer: The MRD analysis was used to help determine the averaging period and is not the core focus of this manuscript. The 15-minute averaging period used here is in-line with other literature in similar landscapes (Reba et al., 2009; Sexstone et al., 2016). Given most sublimation happens during the daytime, we wanted the averaging period to be tailored to the daytime. Therefore, while we investigated both stable and unstable conditions during daytime, we did not run the MRD during nighttime.
Fig 3 – It is difficult to really see trends due to the bar plot, black and blue are hard to distinguish. Can you try curves instead of bars to see if it improves readability (or some other solution)?
Answer: We will change the colors of the black and blue bars to improve readability.
Fig 4(a,b) – The radiative energy flux does not match the sign convention, changing the outward fluxes to -values or everything into absolute values will result in the current plot.
Answer: Thank you very much for catching this. We will adjust these subplots to match the sign convention.
Fig 4(d) – Why is there so much asymmetry in the diurnal cycle of sensible heat flux?
Answer: This diurnal asymmetry in sensible heat flux is common in alpine terrain, with previous studies showing that peak sensible heat flux values are often reached before solar noon, corresponding to a change from down-valley to up-valley winds (Mott et al., 2018; Lehner et al., 2021). Given the scope of our study, however, we have not investigated this behavior in more detail.
References:
Groot Zwaaftink, C. D., Mott, R., and Lehning, M.: Seasonal simulation of drifting snow sublimation in Alpine terrain, Water Resour. Res., 49, 1581–1590, https://doi.org/10.1002/wrcr.20137, 2013.
Lehner, M., Rotach, M. W., Sfyri, E., and Obleitner, F.: Spatial and temporal variations in near-surface energy fluxes in an Alpine valley under synoptically undisturbed and clear-sky conditions, Q. J. Roy. Meteor. Soc., 147, 2173–2196, https://doi.org/10.1002/qj.4016, 2021.
Mott, R., Vionnet, V., and Grünewald, T.: The Seasonal Snow Cover Dynamics: Review on Wind-Driven Coupling Processes, Frontiers in Earth Science, Volume 6 - 2018, https://doi.org/10.3389/feart.2018.00197, 2018.
Reba, M. L., Link, T. E., Marks, D., and Pomeroy, J.: An assessment of corrections for eddy covariance measured turbulent fluxes over snow in mountain environments, Water Resources Research, 45, 0–38, https://doi.org/10.1029/2008WR007045, 2009.
Sexstone, G. A., Clow, D. W., Stannard, D. I., and Fassnacht, S. R.: Comparison of methods for quantifying surface sublimation over seasonally snow-covered terrain, Hydrological Processes, 30, 3373–3389, https://doi.org/10.1002/HYP.10864, 2016.
Citation: https://doi.org/10.5194/egusphere-2026-4268-AC3
-
RC3: 'Comment on egusphere-2026-4268', Anonymous Referee #3, 19 Sep 2026
The paper “Magnitude and controls of snow sublimation at a high-elevation Swiss Alpine site” by Anglin et al. presents an interesting and valuable approach to quantifying snow sublimation using eddy-covariance, snow, and meteorological measurements. The study shows that vapour-pressure gradient (VPG) and wind speed are primary controlling the surface sublimation. The topic is highly relevant, and the combination of observational methods, statistical analyses and machine learing has the potential to provide important insights into snow-atmosphere exchange and is valuable for the community of The Cryosphere.
General comments:
1) The multiresolution decomposition (MRD) analysis is only done with daytime data between 10:00 and 17:00, from which an averaging period of 15 minutes is derived. Could the authors clarify what averaging period would be appropriate if the MRD analysis were performed using the complete dataset? Atmospheric turbulence and flux timescales may differ between daytime and nighttime, particularly under stable conditions. Therefore, an averaging period derived exclusively from daytime data may not be representative of the entire measurement period.
The manuscript states that data gaps occur mainly during nighttime. This may partly be related to the exclusion of nighttime data from the MRD analysis and the resulting choice of averaging period. In addition, stable nighttime conditions may be more likely to fail the stationarity or other quality-control tests. It would therefore be useful if the authors quantified which filtering criteria are primarily responsible for the nighttime data loss.
The amount of missing water vapour flux data following the snow peak also appears substantial with more than 20% (Figure 5). Consequently, the uncertainty associated with the random-forest gap filling is relatively large. Since the comparison between conditions before and after the peak is important to the conclusions of the study, the amount and temporal distribution of gap-filled data require further consideration.
Could the authors investigate whether a sensitivity analysis using modified quality-control thresholds in the preprocessing, particularly for the stationarity criterion, would reduce the data gaps without substantially reducing the data quality? I am not necessarily suggesting that non-stationary data should be included in the final dataset. Rather, such an analysis could help determine whether retaining measured fluxes, potentially with a separate quality flag, leads to lower overall uncertainty than replacing them with modelled values. This would also clarify the trade-off between measurement uncertainty and the uncertainty introduced by gap filling.
2) The vapour pressure gradient (VPG) depends on the estimated snow-surface temperature, assuming saturation at the surface. Capping the surface temperature at 0 °C removes variability whenever the original estimate exceeds this limit. This may weaken its relationship with sublimation and partly explain the increased importance of shortwave radiation after the peak.
The cap also means that VPG variability is largely controlled by the atmospheric vapour pressure measured at 5.4 m. As the conclusions rely strongly on VPG, the authors should demonstrate that the results are robust to uncertainties in the surface temperature estimate and its treatment at 0 °C.
Specific Comments:
Line 153 (Equation4): The paper states that equations for the specific latent heat used to calculate the sublimation flux is provided in the EddyPro 7 documentation. However, I was only able to find the specific latent heat of vaporization and not the latent heat of sublimation, in the documentation. Could the authors provide the relevant reference or clarify where this equation and the value used for the latent heat of sublimation can be found?
Line 312-314: Here it is stated that missing observations mainly occurred during periods when the fluxes were small, such as at night. However, the water-vapour flux record appears to contain a gap-filled period of more than one week at the end of April (Figure 5). Please clarify the cause of this extended data gap and discuss whether the assumption of generally small fluxes is also valid during this period.
Figure 9: Should “(a)” be placed at the beginning of the caption? The description of the temporal evolution appears to refer only to panel (a).
Line 563: Should the initials “RT” be changed to AT, for Adriaan J. Teuling?
Figure A1.: In Figure 5 the light grey shows the gap filled data, while now the grey is supposed to show the raw data. I think between those two “greys” in needs to be distinguished. I suggest using clearly distinguishable colours for raw and gap-filled data to avoid confusion between the figures.
Citation: https://doi.org/10.5194/egusphere-2026-4268-RC3 -
AC4: 'Reply on RC3', Harsh Beria, 02 Oct 2026
Referee #3
Isabella Anglin, on behalf of all authors
The paper “Magnitude and controls of snow sublimation at a high-elevation Swiss Alpine site” by Anglin et al. presents an interesting and valuable approach to quantifying snow sublimation using eddy-covariance, snow, and meteorological measurements. The study shows that vapour-pressure gradient (VPG) and wind speed are primary controlling the surface sublimation. The topic is highly relevant, and the combination of observational methods, statistical analyses and machine learning has the potential to provide important insights into snow-atmosphere exchange and is valuable for the community of The Cryosphere.
We thank referee #3 for a positive evaluation of our study. We have responded to the comments in detail below.
General comments:
1) The multiresolution decomposition (MRD) analysis is only done with daytime data between 10:00 and 17:00, from which an averaging period of 15 minutes is derived. Could the authors clarify what averaging period would be appropriate if the MRD analysis were performed using the complete dataset? Atmospheric turbulence and flux timescales may differ between daytime and nighttime, particularly under stable conditions. Therefore, an averaging period derived exclusively from daytime data may not be representative of the entire measurement period.
The manuscript states that data gaps occur mainly during nighttime. This may partly be related to the exclusion of nighttime data from the MRD analysis and the resulting choice of averaging period. In addition, stable nighttime conditions may be more likely to fail the stationarity or other quality-control tests. It would therefore be useful if the authors quantified which filtering criteria are primarily responsible for the nighttime data loss.
The amount of missing water vapour flux data following the snow peak also appears substantial with more than 20% (Figure 5). Consequently, the uncertainty associated with the random-forest gap filling is relatively large. Since the comparison between conditions before and after the peak is important to the conclusions of the study, the amount and temporal distribution of gap-filled data require further consideration.
Could the authors investigate whether a sensitivity analysis using modified quality-control thresholds in the preprocessing, particularly for the stationarity criterion, would reduce the data gaps without substantially reducing the data quality? I am not necessarily suggesting that non-stationary data should be included in the final dataset. Rather, such an analysis could help determine whether retaining measured fluxes, potentially with a separate quality flag, leads to lower overall uncertainty than replacing them with modelled values. This would also clarify the trade-off between measurement uncertainty and the uncertainty introduced by gap filling.
Answer: We chose to investigate the most suitable averaging time with MRD using daytime data, since this is when we expect the most sublimation to occur. Additionally, this is in-line with previous investigations on surface snow sublimation (Schwat et al., 2025). As running MRD analyses is the most computationally expensive part of the modeling workflow, we hesitate from running full MRD analyses and reprocessing our flux calculations, especially given our averaging periods of 15-minute is very consistent with studies in similar landscapes (Reba et al., 2009; Sexstone et al., 2016).
We will provide more details on the filtering criteria which were responsible for nighttime data loss as well as daytime data loss. Finally, we will also try modifying the quality control thresholds, to evaluate their sensitivities on the results.
2) The vapour pressure gradient (VPG) depends on the estimated snow-surface temperature, assuming saturation at the surface. Capping the surface temperature at 0 °C removes variability whenever the original estimate exceeds this limit. This may weaken its relationship with sublimation and partly explain the increased importance of shortwave radiation after the peak.
The cap also means that VPG variability is largely controlled by the atmospheric vapour pressure measured at 5.4 m. As the conclusions rely strongly on VPG, the authors should demonstrate that the results are robust to uncertainties in the surface temperature estimate and its treatment at 0 °C.
Answer: The snow surface temperature was capped at 0 deg C, because snow surface temperature can physically not exceed this threshold and higher snow surface temperatures are likely sensor artifacts. This approach is also consistent with existing literature (Cristea and Lundquist, 2026). We will evaluate if SWR remains significant when latent heat flux measurements at snow surface temperature of 0 deg C are excluded.
Specific Comments:
Line 153 (Equation4): The paper states that equations for the specific latent heat used to calculate the sublimation flux is provided in the EddyPro 7 documentation. However, I was only able to find the specific latent heat of vaporization and not the latent heat of sublimation, in the documentation. Could the authors provide the relevant reference or clarify where this equation and the value used for the latent heat of sublimation can be found?
Answer: Thank you for this important comment. Our description in the methods may not have been clear enough. Indeed, EddyPro7 only calculates the latent heat of vaporization, which is not suitable for snow surface temperatures below 0 deg C. We therefore calculate the latent heat flux according to Eq. 3 and 4, based on the output of the water vapor flux calculated by EddyPro 7 (l. 149-150). Thus, we will improve the description of the methods to avoid confusing the readers.
Line 312-314: Here it is stated that missing observations mainly occurred during periods when the fluxes were small, such as at night. However, the water-vapour flux record appears to contain a gap-filled period of more than one week at the end of April (Figure 5). Please clarify the cause of this extended data gap and discuss whether the assumption of generally small fluxes is also valid during this period.
Answer: We will report more details about gap lengths and gap-filling in the Methodology section of the revised manuscript. We will additionally investigate the cause of the specific gap at the end of April and provide more information in the revised manuscript.
Figure 9: Should “(a)” be placed at the beginning of the caption? The description of the temporal evolution appears to refer only to panel (a).
Answer: Thank you for catching this. We will adjust this for the final manuscript.
Line 563: Should the initials “RT” be changed to AT, for Adriaan J. Teuling?
Answer: Indeed, this will be adjusted.
Figure A1.: In Figure 5 the light grey shows the gap filled data, while now the grey is supposed to show the raw data. I think between those two “greys” in needs to be distinguished. I suggest using clearly distinguishable colours for raw and gap-filled data to avoid confusion between the figures.
Answer: We will update the figure accordingly.
References:
Cristea, N. C. and Lundquist, J. D.: Sensitivity of Spatial Snowmelt Simulations to Radiative Forcing and Model Top Layer Thickness, Water Resour. Res., 62, e2025WR042968, https://doi.org/10.1029/2025WR042968, 2026.
Reba, M. L., Link, T. E., Marks, D., and Pomeroy, J.: An assessment of corrections for eddy covariance measured turbulent fluxes over snow in mountain environments, Water Resources Research, 45, 0–38, https://doi.org/10.1029/2008WR007045, 2009.
Schwat, E., Hogan, D., Tha, K., Cox, C. J., Butterworth, B. J., Gutmann, E., Vano, J. A., and Lundquist, J. D.: Estimating Snow Sublimation in Complex Terrain: A Season of Intensive Field Measurements and the Role of Vertical Water Vapor Flux Divergence, https://doi.org/10.1175/JHM-D-25-0022.1, 2025
Sexstone, G. A., Clow, D. W., Stannard, D. I., and Fassnacht, S. R.: Comparison of methods for quantifying surface sublimation over seasonally snow-covered terrain, Hydrological Processes, 30, 3373–3389, https://doi.org/10.1002/HYP.10864, 2016.
Citation: https://doi.org/10.5194/egusphere-2026-4268-AC4
-
AC4: 'Reply on RC3', Harsh Beria, 02 Oct 2026
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 282 | 107 | 32 | 421 | 18 | 23 |
- HTML: 282
- PDF: 107
- XML: 32
- Total: 421
- BibTeX: 18
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of egusphere-2026-4268: “Magnitude and controls of snow sublimation at a high-elevation Swiss Alpine site”
I have reviewed the paper “Magnitude and controls of snow sublimation at a high-elevation Swiss Alpine site” by Anglin et al. for publication in The Cryosphere. The study presents results from a field-based experiment at an alpine study site in the European Alps where eddy covariance instrumentation was combined with meteorological and snow monitoring data to quantify surface snow sublimation fluxes over one winter season. Furthermore, the dominant controls on surface sublimation fluxes were assessed using partial correlation and machine learning approaches. Results highlight vapor pressure deficit and wind speed as the dominate drivers of sublimation losses and an average snow season sublimation rate of 0.15 mm day-1 with greater fluxes occurring following peak snow height. I believe this paper addresses an important topic and clearly demonstrates a contribution to the field that is relevant to the readership of The Cryosphere. The paper is clearly organized and well written overall. My comments on the paper are outlined below:
General comments:
1) One limitation of the current study is that only one year of data are available for analysis. So, the natural question is: what would we expect the year-to-year variability of sublimation fluxes to be with respect to the one year of data presented? The authors do a nice job of highlighting results from previous studies in the discussion relevant to interannual variability of sublimation. Furthermore, they put the 2024-25 winter in context from a snow accumulation and snow cover duration perspective. However, given the history of met data collected at this site, did you consider looking at some of the historical data of the identified drivers of sublimation (e.g., VPD, wind speed, radiation) along with snow conditions to provide a historical context? Can the controls of sublimation be used to provide more information about how sublimation fluxes may vary? At a minimum, I think more discussion around this topic and how future research can benefit from quantifying the controls of sublimation would be helpful for the paper (i.e. expand on why it is helpful to identify the controlling variables of sublimation).
2) Recent papers including this one have highlighted well that sublimation fluxes become larger and more important as the snow season progresses with the largest sublimation fluxes occurring presumably during the snowmelt period. It would be helpful to provide snow pit and/or snow scale SWE on your depth graphs to provide more context on when the snowmelt season began for this study. The presence of liquid water in the snowpack and particularly near the surface may be an important condition promoting sublimation. Looking at Figure 2 and Figure A1 it appears that there was still a substantial snowpack on 1 June and the analysis removed most of the active snowmelt period following 1 June. While the authors present clear logic for this period of analysis, it would be helpful to comment on how much more sublimation may have occurred during those last few weeks of snowmelt when substantial sublimation rates may have been occurring. Also, further to this point, Table 2 does not highlight a substantial difference in the controls of sublimation (VPD, wind) between the before peak HS and after peak HS period which I would think is another important point of discussion in the paper.
3) Please comment on how you confirmed that only surface sublimation fluxes were being measured and quantified in this study and that measured fluxes do not include blowing snow sublimation from saltation and/or near surface suspension of blowing snow? Although the authors mention the lower wind speeds at the study site compared to other studies, wind speeds do appear to approach 10 m s-1 (Figure 2). Blowing snow can often obscure the sensor path of IRGASON eddy covariance sensors so it would also be helpful to provide comments on how blowing snow may have contributed to the need to gap filling.
Specific comments:
Line 90: change “fall” to “falls”
Lines 90 – 91: Can you provide more detail or perhaps a wind rose diagram highlighting how common winds from the northwest versus the south are at the site? Which condition is more dominant?
Lines 97 – 99, Table 1: Is the 10 min and 2 min measurement frequency for wind speed, radiation, etc. based on an average of a more frequent scan interval by the datalogger (e.g., 15 seconds), or do the meteorological sensors truly only take one measurement every 10 minutes?
Lines 111 – 113: Data gaps – can you comment more on the frequency and length of data gaps that were required to be filled in this study and if there were periods of missing data where simple linear interpolation was not appropriate and rather implementing and diurnal pattern was needed. Were there any thresholds used for the length of missing periods and what techniques were used?
Lines 142 – 143: How were the 10 minutes met data converted to the 15 minute EC flux interval. Also, later used in the partial correlation and XGBoost analysis?
Equation 4: Does this logic make sense to include in the paper since the snow surface temperature was capped at 0 deg C for the data presented (lines 118 – 120)?
Lines 157 – 158: Please provide more detail on the random forest gap filling approach used in this study. Which variables were used as predictors? Also, either here or in the results section, it is important to note how much gap filling was required over the season and during what time periods. Are there certain meteorological conditions that require gap filling? What about conditions during blowing snow? Although the authors briefly mention gap filling in the discussion, more detail is needed in the methods and results.
Lines 212 – 216: Did you consider using vapor pressure of the atmosphere only in your analysis as a separate predictor? Similarly, did you consider using temperature gradient between the atmosphere and snow surface?
Lines 244 – 249: If you prefer not to present the SSG SWE data, can you plot the SWE observations from snowpits along with your snow depth plots? It is important to try to better highlight potential differences between peak snow height and peak SWE, and more specifically when the snowpack was melting. Also, was liquid water content collected in the snowpits? If so, you could also consider also showing those datasets to further strengthen understanding of when the snowpack was melting and how those periods related to sublimation fluxes.
Lines 269 – 271: Consider if this is the best way to phrase this sentence about the residual term. Although you don’t directly measure internal energy changes, in theory you measure to processes driving these internal energy changes during the snow accumulation period when melt/freeze is not occurring.
Lines 312 – 315: Please include more detail on the gap filled periods.
Figure 5: Consider if there is another way to label cumulative sublimation in the legend and caption to avoid confusion that the plot is presenting only gap filled values.
497 – 503: Surprising to see the Stössel et al., 2010 citation being introduced this late in the discussion considering it was conducted at the same study site. Can results from this previous study be more directly compared to the results presented in this work?
506 – 510: Can you comment on if the unmeasured internal energy changes (i.e. residual term) presented in this study is a realistic magnitude based on melt/freeze cycles and internal snow temperature changes?